Decision Tree Based Power System Dynamic Security Assessment Group 5 Shirley (Xuemeng), Michael (Yu) 3.1.2016 Constraints from previous groups Group Base Case Constraints 1 40GW PG4<501.46 PG65_2>994.13 2 42.5GW PG6>753.61 3 45GW PG6<764.39 47.5GW PG4<418.49 PG9<1743.54 4 5-2 Our Cases Case Load (GW) Load Percentage Change (%) Generation (GW) Gen. Percentage Change (%) 1 48 -4.00 48.79 -4.03 2 49 -2.00 49.81 -2.03 3 (Base) 50 0 50.84 0 4 51 2.00 51.86 2.01 Generator output adjustments Group 4 base case - 48GW (PG6=756MW) 50GW case Scale (PG4=350MW, PG6=760MW) Scale - 49GW (PG6=758MW) - 51GW (PG6=762MW) 3 Bus over voltage issue • Although the power flow of each case successfully converged… • 69 over voltage buses in the base case. • Cause: FIXED shunts. (Shunt capacitors/inductors.) Load scaled while keeping power factor constant Fixed shunts contribute ~ the same amount of reactive power Excessive VAR compensation Over voltage problems (Original case probably has higher load levels) 5-4 TSAT Security Assessment Case No. of secure contingencies No. of insecure contingencies 1 340 16 2 335 21 3 335 21 4 332 24 5-5 Decision tree – unpruned vs pruned Test option – Cross-validation with 10 folds. Unpruned Pruned 5-6 Decision tree – unpruned PG4<=336 5-7 Decision tree –pruned PG4<=336 5-8 Constraints Summary Group Base Case Constraints 1 40GW PG4<501.46 PG65_2>994.13 2 42.5GW PG6>753.61 3 45GW PG6<764.39 4 47.5GW PG4<418.49 PG9<1743.54 5 50GW PG4<336 5-9 Something doesn’t feel right. OH WAIT… 5-10 Our hypothesis • In our study, the four cases are basically scaled from the same case (aside from some minor generation adjustments). • As load grows, the system tends to become more insecure. • For the same contingency, the four cases will have 5 sets of potential security assessment results. 5-11 Our hypothesis Potential outcome 48GW 49GW 50GW 51GW 1 Secure Secure Secure Secure 2 Secure Secure Secure Insecure 3 Secure Secure Insecure Insecure 4 Secure Insecure Insecure Insecure 5 Insecure Insecure Insecure Insecure 5-12 Our hypothesis • All attributes, except FB and OB, are equal in a specific case. • Decision tree first utilizes FB and OB to classify the instances. • For contingencies that have DIFFERENT security assessments in the four cases, the decision tree will attempt to find another attribute to help classify the instances. • Similar trends are seen in the generator output attributes (since they are uniformly scaled), the first attribute (PG4) is selected! Proof: if we switch columns between PG4 and PG6, the decision tree will use PG6! 5-13 Potential issues • This will result in the continuing decreasing limit on Generator 4! • Assume a contingency with the following outcome: Case 1: PG4=336 Case 2: PG4=343 Case 3: PG4=350 Case 4: PG4=357 Secure Insecure Insecure Insecure The decision tree will tend to choose the constraint PG4<=336. This will affect the next group. 5-14 A new set of cases… • In real life system operations, the operating conditions will change in a much more complicated fashion, instead of pure scaling of load and generation. • Based on the previous cases, we altered the output of generator 11. Case 1 2 3 4 PG11 (MW) 2000 1699.64 1734.33 1200 5-15 Decision tree – unpruned vs pruned Test option – Cross-validation with 10 folds. Unpruned Pruned 5-16 Decision tree – unpruned PG4<=336 Limitations? We cannot use the information on generator 4 P5-160>113.71 PG4>350 PG4>336 5-17 SPECIAL THANKS TO: Yongli Thank you for listening. 18